How to decide which settings to use when running Maxent, as well as how to make a bias file

Please consider subscribing to my channel by hitting the "Subscribe" button. It is absolutely free and there are no charges. But if you like what I am doing and you do want to support my channel, consider making a voluntary donation at / joshbanta . It helps me to deliver more content and to keep this channel active. Patreon members get access to ad-free videos and exclusive content.
Exclusive to Patreon paying members: I have created a single R script to automate almost every step of performing species distribution modeling using Maxent. The script even runs Maxent for you. No more running ENMEvaluate separately to find the right model settings, and then inputing them into Maxent. No more hassle creating a bias file and then trying to incorporate it into Maxent on your own (I know a lot of you have trouble with this). See this video for details:
• Run Maxent using R to ...
Want to know how to install the R package ENMEval (used in this tutorial) under Ubuntu Linux? See my other tutorial:
• How to install the R P...

[@jacquelinemattos]

Hi prof. Josh, these are great tutorials and helping me a lot. Thank you very much!

[@irham9321]

Hi prof josh banta, I just want to say thank you for this amazing tutorial. It is very good, and the instructions are clear. Thank you for this tutorial. You are awesome.

[@jamesgeda2787]

Hi prof. first of all i would like to say thank you and please continue this video for interpretation of the result.

[@gregorypitcher7059]

A very clear and useful video for ecologists. Thanks

[@nishanbaral9357]

thank you very much. took my computer exactly 100mins 58sec to run :D

[@sohom-]

Hi! Your tutorial had been a great help so far. Can you provide any reference for using random k fold for more than 50 occurrence points and jackknife for less than 50 occ points? Thanks in advance.

[@danielreis4709]

Great job. I have a doubt about the choice of variables that I will test the correlation. I shouldn't be testing the correlation of the nine variables, right? Which variables do I select to test the correlation and then decide which of these to use in modeling? Example: I want to model a plant. I have the 19 variables available from Wordclin. But as I select within these variables I will test the correlation.

[@user-pe8xq9cd2h]

Thank you for uploading an excellent video. I came to watch this video while studying Maxent. Can I have a couple of questions if you don't mind? Considering that background point is used in Maxent, why is it that this background point should be located around presence point? To my understanding, background point is a point on the hypothesis that background point doesn't generate species, isn't it? If I'm right, I reason that it is right to be located far away from presence point, but it isn't that way and I would like to know why.

[@yadavghimirey]

Hello Josh! Great video. When running the ENMevaluate() function I got the error Error in ENMevaluate(occ, env, bg, tune.args = list(fc = c("L", "LQ", : trying to get slot "errors" from an object of a basic class ("NULL") with no slots. Is this anything to do with rasters I am using? I had quite a few -9999 values in my ascii files.

[@tsydimon]

Hello! Thank you for the video, it's very helpful. I am a newbie in R and I want to ask you something. I got a bias file, but I have a problem with the "color radial gradient" of presence data. As I understood, points of species define the borders of result raster. But if it's so, some data, where color more saturated (and value of presence is greater), I can have a border of this layer. Can I increase the borders of result raster?

[@DrJBanta]

Dear Dmitry, the bias file from my tutorial is made using a kernel density function, which is different from the method you're describing. The kernel density function takes the points and creates a smooth 2-D density surface across the landscape. The bottom line: areas where you sampled more will have a higher density than areas where you sampled less. This is the spatial bias. The algorithm assumes that discovery of locations is proportional to sampling effort. But you will NOT a convex hull polygon with a border, which is what I think you are describing. I hope this helps. -Josh

[@Zudokato9520]

Hi Josh..now in 2022, how could you set the line code to get the enmeval_results like the video with all that rm column. Thanks!! Good video.

[@nisaayob5857]

Hi Josh. Thank you so much for this video, it really helped a lot. Could you perhaps please assist me. I am trying to create a bias file, however I get stuck here Error in ENMevaluate(occ, env, method = "randomkfold", kfolds = 10, categoricals = c(9, : unused argument (kfolds = 10)

[@Nunocesarsa]

i reckon given that now we have R package for maxent many of these hyperparameter tunning methods could be easily implemented into more advanced methods (e.g. bayesian optimization).

[@alexpotter734]

Hi Josh, when trying to create a bias file with my own data I keep getting this error at the rasterize step: "Error in .doCellFromXY(object@ncols, object@nrows, object@extent@xmin, : Not compatible with requested type: [type=character; target=double]." Would you happen to have any idea how I can correct this? Thank you!

[@daphneong6171]

Hi Josh, thank you so much for the informative tutorial. I have a question after running my data (10 occurrence points). I used partitions = "jackknife" instead but it returned with the following: Error in sqrt(np) : non-numeric argument to mathematical function. There are also 2 messages with the following: * Removed 1 occurrence localities that shared the same grid cell. * Removed 8 occurrence points with NA predictor variable values. Not quite sure what error is, I am new to R so any help would be appreciated!

[@DrJBanta]

Dear Daphne, is it possible to send me your files? [e-mail]

[@jillvevers9894]

Hi Josh, within my enmeval results documents it has given me 6 different settings with an AIC of 0. Is there a way to decide which setting is most appropriate? Thanks

[@suzm5626]

Hi Josh, Is there any reason why you chose deltaAICc as the deciding index for selecting which feature to be selected? I see that envevaluate gives other index like average difference in AUC also. Hoping to hear from you

[@DPlusStudents]

See: Warren and Seifert (2011) 'Ecological niche modeling in Maxent: the importance of model complexity and the performance of model selection criteria'

[@carloshenrique-wx3kj]

Hi prof Josh Banta, thank you for this amazing tutorial. I'm trying to run the script and after running > enmeval_results

[@sapnathakur3174]

thank you very much

[@ilhembouderbala5928]

Dear Josh Banta, Thank you for this tutorial it’s very helpful. Actually I have more than 100 species and I want to run maxent to each one of them. My question is: do I have to run ENMeval for each species to extract the bias file or it’s feasible to execute one run for all the species together? Thank you in advance

[@DrJBanta]

You'll have to run it separately for each species I'm afraid. You can automate it using an R script if you're brave enough (requires good R coding skills). See: github.com/jscavetta95/maxentools/

[@abdelmjidboumezzough406]

Salam ilham svp si tu peux me contacter j'ai besoin de votre aide merci bcp ( [e-mail] )

[@joeymarkx1464]

Hi, great tutorials! I have a problem using ENMTools, by including one specific ASCII file, ENMTOOLS give me an error regarding 'illegal devision by 0'. The NoData values in this file are -9999 as is it is in all the others. Cellsize and extent are all the same. Do you have any recommendation of what can be the cause?

[@DrJBanta]

Dear Joey, it sounds like you have been using some of the written (non-video) tutorials on my website (joshbanta.com) as well, such as ENMTools. My experience has been that 95% of the issues such as the one you are experience are caused by this: ***in*** the spreadsheets with your species point locations, you need to make sure the species names are VERY short and contain no spaces or special characters. I am not talking about the names of the files with the occurrence records themselves; I am talking about the names of the species you list in the "species" column within those spreadsheets. Change the names to "species1," "species2", etc., and I bet it will clear up the error.

[@joeymarkx1464]

@@DrJBanta Thanks. The error only occurs when using 1 specific ASCII file. The location file works well, and the species name used is 'Tiger' so without any spaces or so. I was able to run a correlation matrix in R yesterday as well, so have passed this issue.

[@yehtetlwin5468]

Dear professor, When I use length(which...) function for potential background points, it gives 21963 even my occurrence point is only 25. So, should I use 20000 background point for my Maxent modelling? I really hope for your reply.

[@zainulabdin4466]

I hope you found the answer by now. But I think 25 points is not enough for the prensence data itself. Some papers suggest to have a minimum of 300 points for the model to function properly. As for the absence data, there is literature that states that the ideal number for the background points is 10,000 while some other studies suggest that the number of absence must be equal to the number of presence, or if not equal then somewhat close to the each other.

[@ayusmitanaik1762]

I downloaded the maxent software as per your video sir ..but i am not having the maxent..jar file ??Do help sir ...how to do it..

[@ahmadsyihan92]

Hi Josh, I'm encountering this error when running line 57-62; Error in if (any(h

[@hamiddawar8852]

Sir please can you make a video tutorial of SDMtune package.

[@themarinerandthenaturalist]

Hey Josh, great video that was really helpful for me to develop my bias file! I'm wondering why we don't need to normalize/standardize the environmental variables like we usually need to in other modelling processes (such as GLMs)?

[@DrJBanta]

Hello! That is a great question. The best answer I can give is that it is not the convention to standardize environmental variables used by Maxent, so you do not have to do it. Phillips et al. do not standardize the variables in their example dataset, nor do the relevant R packages. So you're on safe ground leaving them alone. I think the machine learning equation simply conforms to whatever the distributions of the variables are like, so I think it's pretty robust to all sorts of distributions and scales. Best, Josh

[@themarinerandthenaturalist]

@@DrJBanta Thanks so much. I just got this question in my masters thesis defence and was stumped beyond the fact that I had never seen it done before or referenced to, so figured the model had to be accounting for this in some way that others don’t. You’re videos were insanely helpful for my whole analysis (modelling predation risk in maxent for livestock kills by snow leopards) and running ENMeval… had to teach myself how to do it from start to finish and just wanted to say thanks! :)

[@thomasbaker26]

Hi Dr. Banta, How would I best cite your ENMevaluate with bias file script? Thanks for all your help!

[@DrJBanta]

Dear Thomas, no need to cite it. You can acknowledge the video in your acknowledgements section if you would like.

[@thomasbaker26]

@@DrJBanta Hi Dr. Banta, Sounds good, thanks!

[@viniciusferracinibissoli6648]

Dear Josh Banta, Thank you for this usuful tutorial. I would like to ask you a question. I have 19 environmental layers, so should I use kfold = 19?

[@DrJBanta]

The number of folds is dependent on the number of occurrence (or presence) records, not the number of environmental layers. Follow the flowchart on my website to decide how many folds to use (www.dropbox.com/s/yw87nxr3nq7v4do/niche%20modeling%20flow%20chart.pdf?dl=0), and then follow the steps in my video (kzbin.info/www/bejne/nGrZqKWQm5uhmbc)

[@viniciusferracinibissoli4792]

Dear Josh, thank you! I visited your website and I tried to check for correlations among environmental layers and remove ones that are highly correlated with each other. When I ran with the data provided in the tutorial I've got the expected results. However, when I ran with my own data I've got "Slope is NaN, intercept is NaN, R is NaN". I tried to change the settings, but I was not sucessful. Do you have any recommendation of what I have done wrong?

[@DrJBanta]

@@viniciusferracinibissoli4792 Hmm, I do not know. One possibility is that the files are too large and ENMTools is giving up. I have another tutorial based on R instead of ENMTools that you could try. Sometimes it doesn't like overly-large files either. www.dropbox.com/s/fyfq1cyku9036u6/raster%20correlations%205-2-17.pdf?dl=0

[@koushikn-bh8cm]

Hii Josh I have a doubt regarding bias files Can we convert our data( I.e .csv format) into .asc format

[@ofer130784]

hi thanks for the tutorial!! would appriciate some help though, Im getting this error: Error in ENMevaluate(occ, env, method = "randomkfold", kfolds = 10, categoricals = c(9, : unused argument (kfolds = 10) and if I change to method = "block" I get this error: Error in ENMevaluate(occ, env, method = "block", categoricals = c(9, 10), : Datasets "occs" and "bg" have different column names. Please make them identical and try again

[@ofer130784]

I changed the column names of bg to be Identical to occ but now I get this error: Error in enm@errors(occs, envs, bg, tune.args, partitions, algorithm, : Maxent settings must include 'rm' (regularization multiplier) and 'fc' (feature class) settings. See ?tune.args for details.

[@erannoor]

@Josh Banta: Thanks for the great video. When I run ENM Evaluate: "enmeval_results

[@DrJBanta]

Hello, you have run out of RAM (memory). This is common. ENMevaluate uses a lot of RAM. You'll need a computer with another 10Gb of RAM, I would say, just to be safe.

[@erannoor]

@@DrJBanta Thank you very much for the reply. I will try with a computer with good specifications.

[@Ammadqazi]

Kindly guide me about some variables, whether they are continuous or categorical? Aspect, slope, elevation, topographic index. Thanks

[@DrJBanta]

Dear Ammad, aspect, slope, and elevation are continuous. I am not sure about topographic index, but it is likely categorical. "Categorical" means that the variable doesn't actually measure anything. Instead it refers to categories. For instance, if I call myself "1" and I call you "2," I am not actually measuring anything. The "1" and the "2" just refer to our names. Similarly, if the topographic index uses "1" to refer to "an area with a lot of hills" and "2" to refer to "an area that is mostly just bottomlands," then it is just giving you names for the categories rather than actually measuring anything. That is what would make it a categorical variable. I hope this helps. Best of luck. -Josh

[@Ammadqazi]

@@DrJBanta Dear Prof Josh, thank you so much for your detailed response. I was confused in some variables, like land cover, topographic index, elevation.

[@DrJBanta]

@@Ammadqazi Land cover is definitely categorical, and topographic index probably is as well. Elevation is a measurement. It measures height above sea level. So elevation is continuous. Best wishes and good luck! -Josh

[@zoegovan1416]

Hi! I am trying to download the maxent software but the link doesn't work, is there anywhere else I can download the data from?

[@DrJBanta]

Dear Zoë, You can download the software here: biodiversityinformatics.amnh.org/open_source/maxent/ I just tried it and it worked. If you are still having problems, wait a few hours and try again. Good luck! Josh

[@sohom-]

The tutorial was very helpful. I used it as well. Ran well, but recently getting an error message like this. Can you help me with this? Error in ENMevaluate(occ, env, method = "randomkfold", kfolds = 10, algorithm = "maxent.jar", : unused argument (kfolds = 10)

[@DrJBanta]

Dear Sohom, it is nothing you are doing wrong. The syntax for the enmevaluate function has changed! I will update the script in the next day or two. Best regards, Josh

[@DrJBanta]

Dear Sohom, I just updated the R scripts. Please re-download them and try again. Good luck! -Josh

[@sohom-]

@@DrJBanta Thanks a lot.

[@thomasbaker26]

Hi Dr. Banta, I am trying to use your updated ENMevaluate script but when I try the final ENMevaluate function, I recieve this error: "Error in data.frame(auc.train = auc.train, cbi.train = cbi.train) :arguments imply differing number of rows: 1, 0" Do you know what would cause this error? The error message is quite cryptic, as I do not know what auc.train and cbi.train are referring to. I assume they are created within the black box-like operations of the ENMevaluate function. Thanks for your help.

[@DrJBanta]

Dear Thomas, have you tried using the older version of the script? In that case, be sure to follow the instructions within the script for installing the OLD version of the 'ENMeval' package. Best, Josh

[@thomasbaker26]

@@DrJBanta Thanks for the response. I originally tried the older version, as I have used it successfully in Spring 2021, but ran into the kfolds error that others have commented on. This prompted me to dig around, discover that ENMeval had been substantially updated since I last used it, and find your new updated script, which I am now using. Do you suggest going back to the old version and trying it again?

[@DrJBanta]

@@thomasbaker26 Yes, I would try that. You are correct about the syntax changing. If you go to my website and re-download the tutorial files, you'll see that there are now two versions: the old script and the new script. The "old" script contains some code to download the old version of the package, so be sure to run those lines of code. Then you should be able to go back to using the old syntax again. Best of luck. Josh.

[@thomasbaker26]

@@DrJBanta Great, thanks!

[@michaelcelone2061]

Hi Josh, I get the following error when I try to generate the background points: Error in .local(.Object, ...) : Do you know what might be causing this? Thanks

[@DrJBanta]

Hmm, that could be a lot of things. First things first: are you sure you set the working directory? Then, are you sure you spelled the names of all of your files correctly when you put them in the script?

[@michaelcelone2061]

@@DrJBanta thanks for your reply. Yes, the script works up until the point where I am selecting background points, i.e. I created the "dens.ras2" raster layer and saved it to my working directory using writeRaster function, and I am able to check the number of potential background points available. Then I am getting the error message when I run the xyFromCell function. Thanks for your help!

[@DrJBanta]

@@michaelcelone2061 Hmm, that makes me think your points file is not set up correctly or contains some hidden characters that are messing it up. I would suggest taking the points file from my tutorial, and just copying-pasting your data into it, and re-saving it. Make sure you don't change the format at all. Fingers crossed!

[@earthaearth9186]

thankyou so much for the tutorial and i am so new to habitat prediction and your vedios helps a lot. but i have a question with my result. i want to use your data for a pratice and i really step-to-step follow your tutorial using your data. however, in my result, the lowest AICc is not LQ0.5(3.332518288 ), but LQ1(0). i really dont know why there is a difference and i also dont know if this is okay or i just made a mistake without noticing. any reply is whole-heartly welcomed. thankyou so much.

[@DrJBanta]

Dear Paris, the lowest AICc can vary, even with the same dataset. Maxent results are a little different every time it runs. So you're all set! You got it to work.

[@earthaearth9186]

Josh Banta THANK YOU SO MUCH🧡🧡🧡🧡🧡🧡🧡🧡🧡🧡🧡🧡🧡🧡🧡🧡🧡🧡🧡🧡🧡🧡🧡🧡

[@earthaearth9186]

Josh Banta so does this mean there won’t be much a difference when I run maxent using LQ0.5 or LQ1?

[@DrJBanta]

@@earthaearth9186 yes I think that's right

[@Amitsingh303]

Dear Josh Banta can this method also applied for 60 occurrence points

[@DrJBanta]

Certainly. You would use the cross-validation method in the flow chart.

[@Philipecorcino]

Hi Josh!! I'm doing a worldwide modeling and when running your script at the time I'm going to run ENMeval, with 90% of the process I get the following message: Error: cannot allocate vector of size 122.1 Mb. The problem would be the worldwide extent of Bios?

[@DrJBanta]

You are correct, you've run out of RAM. Find a machine with more RAM or else make your spatial extent smaller for your layers.

[@Philipecorcino]

@@DrJBanta Thanks a lot and congrats for your tutorials!

[@SHIVANISINGH-yy8um]

When we run maxent with the occurrences given in the tutorial(occ.csv) , it gave results. But when we are trying to run with our occurrences, it is loading the file into occ but when we click plot, nothing is being plotted on the graph. No. of occurrence points we have is 5,060. Kindly answer.

[@sreechandravaranasi6221]

Josh banta kindly answer even I have the same problem

[@ryeo1998]

ha even i have the same problem

[@DrJBanta]

Dear Shivani, It is hard for me to diagnose the problem without seeing your files, but I strongly suspect that either: (1) Your points (occurrence) data are not in the projection as your environmental layers. (2) Your points data lies outside of the spatial extent of your environmental layers. To clarify: my environmental layers will not work with your occurrence points. You need to clip your own environmental layers after downloading them fresh from the source databases (like WorldClim). This is because your points data and my environmental layers are very likely covering different spatial extents. My environmental layers only cover Texas, for instance. I hope this helps. Josh

[@SHIVANISINGH-yy8um]

@@DrJBanta Thankyou so much,we could rectify the error, your video was of great help.

[@ishaqalishahdurrani8167]

Hey dear professor... I downloaded enmevaluate new files today. It's R script opens empty in R studio... Can you please help me out..?? Thanks in anticipation

[@DrJBanta]

Dear Ishaq, I just tried downloading the files, and they are not empty. Are you sure you opened the files from the correct folder on your computer? Try opening the files in a text editor to see if they appear empty there as well. Best of luck, Josh

[@ishaqalishahdurrani8167]

@@DrJBanta Bundle of thanks dear Professor... I restarted my system and now it works. Your tutorials and supplements are unparallel... Great contribution to scientific community...👍👍💓

[@ishaqalishahdurrani8167]

@@DrJBanta Dear professor, One another question please. How can I get biocategorical variables please for my study area?

[@emilymcculloch-jones8799]

Hi Josh, I run your Tut as is and when I try to run the ENMevaluate(), as you have it, I get this error: Error in ENMevaluate(occ, env, method = "randomkfold", kfolds = 10, categoricals = c(9, : unused argument (kfolds = 10) If I remove the kfolds argument, I continue to get multiple other errors. Any advice?

[@DrJBanta]

Dear Emily, please follow up with me via email and we can work on it.

[@irham9321]

@@DrJBanta i also have the same problem with this unused argument. Can i email you about this?

[@DrJBanta]

@@irham9321 Dear Irham, Please see the "hashed out" comments at the beginning of the script. You need to download an older version of the ENMEval package for now. Run that part of the script (without the has symbol) and you should be good to go.

[@irham9321]

@@DrJBanta i just saw your comment that you have updated the r script. After i comment here. And it worked! One more question. Is the k method suitable for species that are not good with dispersion? Im trying to do maxent on tarantula.

[@DrJBanta]

@@irham9321 Can you remind me what you mean by the K method?

[@vanuhihambardzumyan1579]

Hi Josh, thank you amazing tutorials, I am following them until this point, but I have an error here which I can't solve, I reeely hope someone can help me here: So I have only 15 occurrence points, that's why I use: enmeval_results

[@DrJBanta]

Dear Vanuhi, the problem is that the script you are using is incompatible with the version of the R package enmeval that you are using. To fix the problem: re-visit my webpage and re-download the tutorial files. Be sure to download the OLD version of the files. Then, pay careful attention to the "hashed-out" comments at the beginning of the script that provide important instructions for installing the old version of the enmeval package to use with that script. I hope this helps. Best, Josh

[@sohom-]

Sorry for coming up with another error, if you may please help. > bg

[@DrJBanta]

Dear Sohom, it sounds like you don't have enough background points available. Change the 10000 in the script to 5000 and try again. Best regards, Josh

[@sohom-]

@@DrJBanta I checked maximum background points. Its more than 10000

[@DrJBanta]

@@sohom- Dear Sohom, is the tutorial working on the sample data? I would always start there to make sure it's not a problem with my code or with the sample data. Best, Josh

[@sohom-]

@@DrJBantaYeah, its worked with sample data. Even I ran some models with my own data last week. They worked fine as well. May be there are changes in the package this week. Can you explain what these 'size' and 'replace' in the error message mean?. Thanks in advance.

[@DrJBanta]

@@sohom- Dear Sohom, you are correct that the syntax of the ENMEvaluate() function has changed. I have made an updated script on my website. Re-download it and read the beginning of the script carefully. Good luck! Josh